AVOCADawg

Woof woof wuf wuff ruff w00f wooF woof!Woof woof wuf wuff ruff w00f wooF woof!Woof woof wuf wuff ruff w00f wooF woof!

Inspiration

We were inspried by the recent rise in autopilots and the use of state of the art technologies such as Single Shot Detector Multibox models. Thus we integrated different knowledges such as object detection with neural network and color-based object tracking to build an autnonomous RC vehicle.

What it does

The prototype is able to detect a tennis ball by hue saturation value, and is able to drive on its own to the object that it is tracking. It is also able to detect any stop signs and slow down/brake upon detection. Furthermore, the prototype project includes an app that can feed livestream video of the webcam the vehicle is using.

How we built it

We used an off the shelf RC car and mounted a Nvidia tx2 Jetson for main computing ability. ROS with c++ has been used for building the main controls and the communications, and we used python along with the Tensorflow object detection API to build the main script to run the car and to detect the stop signs.

Challenges I ran into

Our challenges comprised of both hardware and software challenges, where the biggest initial challenge took place when we were configuring the setup of the car. On the other hand working on ROS and python to run the car was a lot of difficulty to develop and integrate altogether.

Accomplishments that I'm proud of

We are proud to present a prototype that can successfully track a tennis ball and drive with an autopilot, alongside with stop-sign detection and its response to it.

What I learned

We learned a lot during this project, including but definitely not limited to implementing ROS, applying Tensorflow API to a real life application and on hardware, and integrating the entire project's modules.

What's next for Fetch

We hope to build much more robust models by training our SSD model at much longer duration and collect more diverse data for improvement of current detection and for other possible uses e.g track.

Built With

• cpp
• python
• ros
• tensorflow